backlog-refinementlisted
Install: claude install-skill singleton-sd/ai-plattform-skills
# Backlog Refinement
Use this skill to turn a rough idea, transcript, feature request, bug, or existing ticket into work that can be executed without hidden product decisions.
For multi-ticket feature planning, also apply `idea-to-delivery`.
## First decide the shape
- Tiny isolated change: one implementation task.
- Medium feature: Epic/initiative + roughly 2–5 implementation tasks.
- Large/cross-cutting or materially uncertain work: discovery/design first, then Epic + implementation slices.
- Manual requirement: explicit human-operation task rather than an implementation task an agent cannot finish.
Default slicing rule: one independently mergeable PR is usually one implementation task.
## Refined item format
```text
Title: <concise action-oriented title>
Area: <project-defined technical/product surface>
Pillar: <project-defined domain capability or None>
Work Type: Feature | Bug | Technical Debt | Discovery | Infrastructure | Documentation | Human Action
Execution: AI | Human | AI + Human
Parent Epic: <title/link or None>
Goal:
<observable outcome>
User story (optional):
As a [persona], I want [action] so that [outcome].
Context:
<why this matters>
Scope:
- <included work>
Acceptance criteria:
- [ ] <specific testable condition>
Technical direction:
- <constraints, patterns, likely repo paths when known>
Preview / seed scenarios:
- <how user-facing behavior or a bug is demonstrated when applicable>
Testing:
- <automated expectations>
- <human validation exp